Speaking on stage last year about the rise of artificial intelligence on devices, Arm Holdings CEO Simon Segars touted the capabilities of the chip giant’s Cortex-M4 processor to help identify objects using image recognition technology. But the circuit board on the slide behind him showed another logo: Xnor.ai.
Not many people knew that name at the time, even inside the industry, but they do now. Xnor.ai is the Seattle-based startup that was acquired by Apple recently for around $200 million, as first reported by GeekWire last week.
Its inclusion in the Arm CEO’s presentation illustrated the growing attention for Xnor.ai at the time, fueled by the industry’s rising fascination with artificial intelligence on “the edge,” in the small cameras, sensors, and many other types of devices proliferating around the world.
“You need to push intelligence to the edge,” explained Alexei Andreev, co-founder and managing director at Autotech Ventures, which was an early investor in Xnor.ai. “You cannot move data in and out to the data centers anymore if you want to make fast decisions, if you don’t want to overload your data networks.”
Xnor.ai’s capabilities are a big part of what makes Apple’s acquisition of the company so intriguing.
Based on technology developed at the University of Washington, and incubated at the Allen Institute for Artificial Intelligence (AI2), the startup streamlined and simplified the calculations needed for image recognition and other components of artificial intelligence. Devices using Xnor.ai’s technology, for example, are able to identify objects such as people, backpacks and vehicles without needing to rely on a network connection to a data center, or a device’s resource-intensive graphics processing unit. Instead, they can use a CPU on devices.
“You just can’t afford to take all of the data to where the processing can be done,” said Arm’s Segars at a separate event. “You need to take the processing to the data.”
Andreev and others we spoke with for this story were careful to say that they don’t know exactly what Apple plans to do with Xnor.ai’s technology.
Before news of the deal broke, Apple appeared determined to keep even the fact that it acquired Xnor.ai a secret, creating a company called Xylophone Capital Corp. to complete the transaction, according to Delaware corporations filings obtained by GeekWire in the course of reporting the story. Xnor.ai’s full website was also taken taken offline in recent weeks, replaced with a bare-bones placeholder page.
But the nature of the Xnor.ai technology suggests a few areas where Apple’s acquisition of Xnor.ai could come into play.
- Better battery life: The ability to implement AI without using a GPU could translate into longer battery life on existing Apple devices such as the iPhone, Apple Watch, iPad and others, while improving the company’s Siri virtual assistant.
- More private and secure: The ability to keep images and other data on device, without transferring over a network connection to the cloud, reduces the chances of privacy leaks or hacks.
- Developer tools: Xnor.ai offers a platform for developers to implement AI in their apps, using its on-device approach. This could become the basis of new Apple developer tools.
- New products: Apart from implementing Xnor.ai’s technology in its existing products, the ability to conduct AI in low-powered devices raises the possibility of entirely new Apple products.
In one example of the technology’s potential, Xnor last year demonstrated the ability to run its AI technology on a solar-powered AI chip that could run without other power for more than 30 years. “To us, this is as big as when somebody invented a light bulb,” Xnor.ai’s co-founder, Ali Farhadi, said the time.
The key to Xnor’s edge technology is a chip that can run AI software on mere milliwatts of power, so that a coin-sized battery could theoretically keep the chip running for 30 years. The system takes advantage of low-power wireless technologies such as Narrowband IoT and LoRa.
Madrona Venture Group and AI2 put $2.6 million in seed funding into Xnor.ai in 2017, and Madrona led a $12 million Series A financing round in 2018 with additional backing from Autotech, NGP Capital and Catapult Ventures. When the venture was spun out, it had just six employees, but that has since grown to about 70 people.
Apple has made 20 artificial intelligence acquisitions in the last decade, more than any other tech company, and it has sought to position itself as a bastion of user privacy and security. The company acquired another company specializing in on-device AI, Silk Labs, in November 2018.
In 2016, Apple acquired Turi, another Seattle startup specializing in machine learning and AI, which also has roots at the University of Washington, and for a similar price, about $200 million. Apple has since released developer tools based on Turi’s platform and incorporated Turi’s technology into its own products.
However, Apple wasn’t the only company knocking on Xnor.ai’s door. Sources told GeekWire last week that Intel and Amazon had formal discussions with Xnor.ai about a possible acquisition. The Financial Times reported that Microsoft also approached the company.
GeekWire’s Alan Boyle, Taylor Soper and Kevin Lisota contributed to this report.